We present Random Partition Relaxation (RPR), a method for strong quantization of neural networks weight to binary (+1/-1) and ternary (+1/0/-1) values. Starting from a pre-trained model, we quantize the weights and then relax random partitions of them to their continuous values for retraining before re-quantizing them and switching to another weight partition for further adaptation. We demonstrate binary and ternary-weight networks with accuracies beyond the state-of-the-art for GoogLeNet and competitive performance for ResNet-18 and ResNet-50 using an SGD-based training method that can easily be integrated into existing frameworks.
@article{arxiv.2001.01091,
title = {RPR: Random Partition Relaxation for Training; Binary and Ternary Weight Neural Networks},
author = {Lukas Cavigelli and Luca Benini},
journal= {arXiv preprint arXiv:2001.01091},
year = {2020}
}